From a0d13aac7022f95ec161c18d18e3d81172666ed8 Mon Sep 17 00:00:00 2001 From: Wedson Almeida Filho Date: Thu, 22 Aug 2024 16:37:53 +0000 Subject: rust: rbtree: add red-black tree implementation backed by the C version The rust rbtree exposes a map-like interface over keys and values, backed by the kernel red-black tree implementation. Values can be inserted, deleted, and retrieved from a `RBTree` by key. This base abstraction is used by binder to store key/value pairs and perform lookups, for example the patch "[PATCH RFC 03/20] rust_binder: add threading support" in the binder RFC [1]. Link: https://lore.kernel.org/rust-for-linux/20231101-rust-binder-v1-3-08ba9197f637@google.com/ [1] Signed-off-by: Wedson Almeida Filho Reviewed-by: Alice Ryhl Tested-by: Alice Ryhl Reviewed-by: Boqun Feng Reviewed-by: Benno Lossin Signed-off-by: Matt Gilbride Link: https://lore.kernel.org/r/20240822-b4-rbtree-v12-1-014561758a57@google.com [ Updated link to docs.kernel.org. - Miguel ] Signed-off-by: Miguel Ojeda --- rust/helpers/helpers.c | 1 + rust/helpers/rbtree.c | 9 ++ rust/kernel/lib.rs | 1 + rust/kernel/rbtree.rs | 432 +++++++++++++++++++++++++++++++++++++++++++++++++ 4 files changed, 443 insertions(+) create mode 100644 rust/helpers/rbtree.c create mode 100644 rust/kernel/rbtree.rs (limited to 'rust') diff --git a/rust/helpers/helpers.c b/rust/helpers/helpers.c index 173533616c91..30f40149f3a9 100644 --- a/rust/helpers/helpers.c +++ b/rust/helpers/helpers.c @@ -15,6 +15,7 @@ #include "kunit.c" #include "mutex.c" #include "page.c" +#include "rbtree.c" #include "refcount.c" #include "signal.c" #include "slab.c" diff --git a/rust/helpers/rbtree.c b/rust/helpers/rbtree.c new file mode 100644 index 000000000000..6d404b84a9b5 --- /dev/null +++ b/rust/helpers/rbtree.c @@ -0,0 +1,9 @@ +// SPDX-License-Identifier: GPL-2.0 + +#include + +void rust_helper_rb_link_node(struct rb_node *node, struct rb_node *parent, + struct rb_node **rb_link) +{ + rb_link_node(node, parent, rb_link); +} diff --git a/rust/kernel/lib.rs b/rust/kernel/lib.rs index 9baea9e9ee1a..f10b06a78b9d 100644 --- a/rust/kernel/lib.rs +++ b/rust/kernel/lib.rs @@ -44,6 +44,7 @@ pub mod net; pub mod page; pub mod prelude; pub mod print; +pub mod rbtree; mod static_assert; #[doc(hidden)] pub mod std_vendor; diff --git a/rust/kernel/rbtree.rs b/rust/kernel/rbtree.rs new file mode 100644 index 000000000000..cf25437c795f --- /dev/null +++ b/rust/kernel/rbtree.rs @@ -0,0 +1,432 @@ +// SPDX-License-Identifier: GPL-2.0 + +//! Red-black trees. +//! +//! C header: [`include/linux/rbtree.h`](srctree/include/linux/rbtree.h) +//! +//! Reference: + +use crate::{alloc::Flags, bindings, container_of, error::Result, prelude::*}; +use alloc::boxed::Box; +use core::{ + cmp::{Ord, Ordering}, + marker::PhantomData, + mem::MaybeUninit, + ptr::{addr_of_mut, NonNull}, +}; + +/// A red-black tree with owned nodes. +/// +/// It is backed by the kernel C red-black trees. +/// +/// # Examples +/// +/// In the example below we do several operations on a tree. We note that insertions may fail if +/// the system is out of memory. +/// +/// ``` +/// use kernel::{alloc::flags, rbtree::{RBTree, RBTreeNode, RBTreeNodeReservation}}; +/// +/// // Create a new tree. +/// let mut tree = RBTree::new(); +/// +/// // Insert three elements. +/// tree.try_create_and_insert(20, 200, flags::GFP_KERNEL)?; +/// tree.try_create_and_insert(10, 100, flags::GFP_KERNEL)?; +/// tree.try_create_and_insert(30, 300, flags::GFP_KERNEL)?; +/// +/// // Check the nodes we just inserted. +/// { +/// assert_eq!(tree.get(&10).unwrap(), &100); +/// assert_eq!(tree.get(&20).unwrap(), &200); +/// assert_eq!(tree.get(&30).unwrap(), &300); +/// } +/// +/// // Replace one of the elements. +/// tree.try_create_and_insert(10, 1000, flags::GFP_KERNEL)?; +/// +/// // Check that the tree reflects the replacement. +/// { +/// assert_eq!(tree.get(&10).unwrap(), &1000); +/// assert_eq!(tree.get(&20).unwrap(), &200); +/// assert_eq!(tree.get(&30).unwrap(), &300); +/// } +/// +/// // Change the value of one of the elements. +/// *tree.get_mut(&30).unwrap() = 3000; +/// +/// // Check that the tree reflects the update. +/// { +/// assert_eq!(tree.get(&10).unwrap(), &1000); +/// assert_eq!(tree.get(&20).unwrap(), &200); +/// assert_eq!(tree.get(&30).unwrap(), &3000); +/// } +/// +/// // Remove an element. +/// tree.remove(&10); +/// +/// // Check that the tree reflects the removal. +/// { +/// assert_eq!(tree.get(&10), None); +/// assert_eq!(tree.get(&20).unwrap(), &200); +/// assert_eq!(tree.get(&30).unwrap(), &3000); +/// } +/// +/// # Ok::<(), Error>(()) +/// ``` +/// +/// In the example below, we first allocate a node, acquire a spinlock, then insert the node into +/// the tree. This is useful when the insertion context does not allow sleeping, for example, when +/// holding a spinlock. +/// +/// ``` +/// use kernel::{alloc::flags, rbtree::{RBTree, RBTreeNode}, sync::SpinLock}; +/// +/// fn insert_test(tree: &SpinLock>) -> Result { +/// // Pre-allocate node. This may fail (as it allocates memory). +/// let node = RBTreeNode::new(10, 100, flags::GFP_KERNEL)?; +/// +/// // Insert node while holding the lock. It is guaranteed to succeed with no allocation +/// // attempts. +/// let mut guard = tree.lock(); +/// guard.insert(node); +/// Ok(()) +/// } +/// ``` +/// +/// In the example below, we reuse an existing node allocation from an element we removed. +/// +/// ``` +/// use kernel::{alloc::flags, rbtree::{RBTree, RBTreeNodeReservation}}; +/// +/// // Create a new tree. +/// let mut tree = RBTree::new(); +/// +/// // Insert three elements. +/// tree.try_create_and_insert(20, 200, flags::GFP_KERNEL)?; +/// tree.try_create_and_insert(10, 100, flags::GFP_KERNEL)?; +/// tree.try_create_and_insert(30, 300, flags::GFP_KERNEL)?; +/// +/// // Check the nodes we just inserted. +/// { +/// assert_eq!(tree.get(&10).unwrap(), &100); +/// assert_eq!(tree.get(&20).unwrap(), &200); +/// assert_eq!(tree.get(&30).unwrap(), &300); +/// } +/// +/// // Remove a node, getting back ownership of it. +/// let existing = tree.remove(&30).unwrap(); +/// +/// // Check that the tree reflects the removal. +/// { +/// assert_eq!(tree.get(&10).unwrap(), &100); +/// assert_eq!(tree.get(&20).unwrap(), &200); +/// assert_eq!(tree.get(&30), None); +/// } +/// +/// // Create a preallocated reservation that we can re-use later. +/// let reservation = RBTreeNodeReservation::new(flags::GFP_KERNEL)?; +/// +/// // Insert a new node into the tree, reusing the previous allocation. This is guaranteed to +/// // succeed (no memory allocations). +/// tree.insert(reservation.into_node(15, 150)); +/// +/// // Check that the tree reflect the new insertion. +/// { +/// assert_eq!(tree.get(&10).unwrap(), &100); +/// assert_eq!(tree.get(&15).unwrap(), &150); +/// assert_eq!(tree.get(&20).unwrap(), &200); +/// } +/// +/// # Ok::<(), Error>(()) +/// ``` +/// +/// # Invariants +/// +/// Non-null parent/children pointers stored in instances of the `rb_node` C struct are always +/// valid, and pointing to a field of our internal representation of a node. +pub struct RBTree { + root: bindings::rb_root, + _p: PhantomData>, +} + +// SAFETY: An [`RBTree`] allows the same kinds of access to its values that a struct allows to its +// fields, so we use the same Send condition as would be used for a struct with K and V fields. +unsafe impl Send for RBTree {} + +// SAFETY: An [`RBTree`] allows the same kinds of access to its values that a struct allows to its +// fields, so we use the same Sync condition as would be used for a struct with K and V fields. +unsafe impl Sync for RBTree {} + +impl RBTree { + /// Creates a new and empty tree. + pub fn new() -> Self { + Self { + // INVARIANT: There are no nodes in the tree, so the invariant holds vacuously. + root: bindings::rb_root::default(), + _p: PhantomData, + } + } +} + +impl RBTree +where + K: Ord, +{ + /// Tries to insert a new value into the tree. + /// + /// It overwrites a node if one already exists with the same key and returns it (containing the + /// key/value pair). Returns [`None`] if a node with the same key didn't already exist. + /// + /// Returns an error if it cannot allocate memory for the new node. + pub fn try_create_and_insert( + &mut self, + key: K, + value: V, + flags: Flags, + ) -> Result>> { + Ok(self.insert(RBTreeNode::new(key, value, flags)?)) + } + + /// Inserts a new node into the tree. + /// + /// It overwrites a node if one already exists with the same key and returns it (containing the + /// key/value pair). Returns [`None`] if a node with the same key didn't already exist. + /// + /// This function always succeeds. + pub fn insert(&mut self, RBTreeNode { node }: RBTreeNode) -> Option> { + let node = Box::into_raw(node); + // SAFETY: `node` is valid at least until we call `Box::from_raw`, which only happens when + // the node is removed or replaced. + let node_links = unsafe { addr_of_mut!((*node).links) }; + + // The parameters of `bindings::rb_link_node` are as follows: + // - `node`: A pointer to an uninitialized node being inserted. + // - `parent`: A pointer to an existing node in the tree. One of its child pointers must be + // null, and `node` will become a child of `parent` by replacing that child pointer + // with a pointer to `node`. + // - `rb_link`: A pointer to either the left-child or right-child field of `parent`. This + // specifies which child of `parent` should hold `node` after this call. The + // value of `*rb_link` must be null before the call to `rb_link_node`. If the + // red/black tree is empty, then it’s also possible for `parent` to be null. In + // this case, `rb_link` is a pointer to the `root` field of the red/black tree. + // + // We will traverse the tree looking for a node that has a null pointer as its child, + // representing an empty subtree where we can insert our new node. We need to make sure + // that we preserve the ordering of the nodes in the tree. In each iteration of the loop + // we store `parent` and `child_field_of_parent`, and the new `node` will go somewhere + // in the subtree of `parent` that `child_field_of_parent` points at. Once + // we find an empty subtree, we can insert the new node using `rb_link_node`. + let mut parent = core::ptr::null_mut(); + let mut child_field_of_parent: &mut *mut bindings::rb_node = &mut self.root.rb_node; + while !child_field_of_parent.is_null() { + parent = *child_field_of_parent; + + // We need to determine whether `node` should be the left or right child of `parent`, + // so we will compare with the `key` field of `parent` a.k.a. `this` below. + // + // SAFETY: By the type invariant of `Self`, all non-null `rb_node` pointers stored in `self` + // point to the links field of `Node` objects. + let this = unsafe { container_of!(parent, Node, links) }; + + // SAFETY: `this` is a non-null node so it is valid by the type invariants. `node` is + // valid until the node is removed. + match unsafe { (*node).key.cmp(&(*this).key) } { + // We would like `node` to be the left child of `parent`. Move to this child to check + // whether we can use it, or continue searching, at the next iteration. + // + // SAFETY: `parent` is a non-null node so it is valid by the type invariants. + Ordering::Less => child_field_of_parent = unsafe { &mut (*parent).rb_left }, + // We would like `node` to be the right child of `parent`. Move to this child to check + // whether we can use it, or continue searching, at the next iteration. + // + // SAFETY: `parent` is a non-null node so it is valid by the type invariants. + Ordering::Greater => child_field_of_parent = unsafe { &mut (*parent).rb_right }, + Ordering::Equal => { + // There is an existing node in the tree with this key, and that node is + // `parent`. Thus, we are replacing parent with a new node. + // + // INVARIANT: We are replacing an existing node with a new one, which is valid. + // It remains valid because we "forgot" it with `Box::into_raw`. + // SAFETY: All pointers are non-null and valid. + unsafe { bindings::rb_replace_node(parent, node_links, &mut self.root) }; + + // INVARIANT: The node is being returned and the caller may free it, however, + // it was removed from the tree. So the invariants still hold. + return Some(RBTreeNode { + // SAFETY: `this` was a node in the tree, so it is valid. + node: unsafe { Box::from_raw(this.cast_mut()) }, + }); + } + } + } + + // INVARIANT: We are linking in a new node, which is valid. It remains valid because we + // "forgot" it with `Box::into_raw`. + // SAFETY: All pointers are non-null and valid (`*child_field_of_parent` is null, but `child_field_of_parent` is a + // mutable reference). + unsafe { bindings::rb_link_node(node_links, parent, child_field_of_parent) }; + + // SAFETY: All pointers are valid. `node` has just been inserted into the tree. + unsafe { bindings::rb_insert_color(node_links, &mut self.root) }; + None + } + + /// Returns a node with the given key, if one exists. + fn find(&self, key: &K) -> Option>> { + let mut node = self.root.rb_node; + while !node.is_null() { + // SAFETY: By the type invariant of `Self`, all non-null `rb_node` pointers stored in `self` + // point to the links field of `Node` objects. + let this = unsafe { container_of!(node, Node, links) }; + // SAFETY: `this` is a non-null node so it is valid by the type invariants. + node = match key.cmp(unsafe { &(*this).key }) { + // SAFETY: `node` is a non-null node so it is valid by the type invariants. + Ordering::Less => unsafe { (*node).rb_left }, + // SAFETY: `node` is a non-null node so it is valid by the type invariants. + Ordering::Greater => unsafe { (*node).rb_right }, + Ordering::Equal => return NonNull::new(this.cast_mut()), + } + } + None + } + + /// Returns a reference to the value corresponding to the key. + pub fn get(&self, key: &K) -> Option<&V> { + // SAFETY: The `find` return value is a node in the tree, so it is valid. + self.find(key).map(|node| unsafe { &node.as_ref().value }) + } + + /// Returns a mutable reference to the value corresponding to the key. + pub fn get_mut(&mut self, key: &K) -> Option<&mut V> { + // SAFETY: The `find` return value is a node in the tree, so it is valid. + self.find(key) + .map(|mut node| unsafe { &mut node.as_mut().value }) + } + + /// Removes the node with the given key from the tree. + /// + /// It returns the node that was removed if one exists, or [`None`] otherwise. + fn remove_node(&mut self, key: &K) -> Option> { + let mut node = self.find(key)?; + + // SAFETY: The `find` return value is a node in the tree, so it is valid. + unsafe { bindings::rb_erase(&mut node.as_mut().links, &mut self.root) }; + + // INVARIANT: The node is being returned and the caller may free it, however, it was + // removed from the tree. So the invariants still hold. + Some(RBTreeNode { + // SAFETY: The `find` return value was a node in the tree, so it is valid. + node: unsafe { Box::from_raw(node.as_ptr()) }, + }) + } + + /// Removes the node with the given key from the tree. + /// + /// It returns the value that was removed if one exists, or [`None`] otherwise. + pub fn remove(&mut self, key: &K) -> Option { + self.remove_node(key).map(|node| node.node.value) + } +} + +impl Default for RBTree { + fn default() -> Self { + Self::new() + } +} + +impl Drop for RBTree { + fn drop(&mut self) { + // SAFETY: `root` is valid as it's embedded in `self` and we have a valid `self`. + let mut next = unsafe { bindings::rb_first_postorder(&self.root) }; + + // INVARIANT: The loop invariant is that all tree nodes from `next` in postorder are valid. + while !next.is_null() { + // SAFETY: All links fields we create are in a `Node`. + let this = unsafe { container_of!(next, Node, links) }; + + // Find out what the next node is before disposing of the current one. + // SAFETY: `next` and all nodes in postorder are still valid. + next = unsafe { bindings::rb_next_postorder(next) }; + + // INVARIANT: This is the destructor, so we break the type invariant during clean-up, + // but it is not observable. The loop invariant is still maintained. + + // SAFETY: `this` is valid per the loop invariant. + unsafe { drop(Box::from_raw(this.cast_mut())) }; + } + } +} + +/// A memory reservation for a red-black tree node. +/// +/// +/// It contains the memory needed to hold a node that can be inserted into a red-black tree. One +/// can be obtained by directly allocating it ([`RBTreeNodeReservation::new`]). +pub struct RBTreeNodeReservation { + node: Box>>, +} + +impl RBTreeNodeReservation { + /// Allocates memory for a node to be eventually initialised and inserted into the tree via a + /// call to [`RBTree::insert`]. + pub fn new(flags: Flags) -> Result> { + Ok(RBTreeNodeReservation { + node: as BoxExt<_>>::new_uninit(flags)?, + }) + } +} + +// SAFETY: This doesn't actually contain K or V, and is just a memory allocation. Those can always +// be moved across threads. +unsafe impl Send for RBTreeNodeReservation {} + +// SAFETY: This doesn't actually contain K or V, and is just a memory allocation. +unsafe impl Sync for RBTreeNodeReservation {} + +impl RBTreeNodeReservation { + /// Initialises a node reservation. + /// + /// It then becomes an [`RBTreeNode`] that can be inserted into a tree. + pub fn into_node(self, key: K, value: V) -> RBTreeNode { + let node = Box::write( + self.node, + Node { + key, + value, + links: bindings::rb_node::default(), + }, + ); + RBTreeNode { node } + } +} + +/// A red-black tree node. +/// +/// The node is fully initialised (with key and value) and can be inserted into a tree without any +/// extra allocations or failure paths. +pub struct RBTreeNode { + node: Box>, +} + +impl RBTreeNode { + /// Allocates and initialises a node that can be inserted into the tree via + /// [`RBTree::insert`]. + pub fn new(key: K, value: V, flags: Flags) -> Result> { + Ok(RBTreeNodeReservation::new(flags)?.into_node(key, value)) + } +} + +// SAFETY: If K and V can be sent across threads, then it's also okay to send [`RBTreeNode`] across +// threads. +unsafe impl Send for RBTreeNode {} + +// SAFETY: If K and V can be accessed without synchronization, then it's also okay to access +// [`RBTreeNode`] without synchronization. +unsafe impl Sync for RBTreeNode {} + +struct Node { + links: bindings::rb_node, + key: K, + value: V, +} -- cgit v1.2.3